基于R语言的多目标优化甘草提取物工艺研究

Multi-level optimal extraction technique of licorice using R language application

目的优化同时提取甘草中皂苷和总黄酮的工艺条件。方法在单因素试验的基础上,氨浓度(A)、乙醇浓度(B)、回流时间(C)、液料比(D)为自变量,选用响应面法中的中心组合设计(CCD)进行4因素5水平试验,采用紫外分光光度法进行测定,以甘草中皂苷和总黄酮含量作为检测指标,检测波长分别为252 nm和510 nm。运用R语言环境下的熵权法对上述2指标进行权重赋值,建立3层结构的BP神经网络模型,对不同隐层神经元(size)的个数进行模型的测试训练,最后采用R语言的实数编码程序,建立并优化遗传算法数学模型,对提取工艺进行目标寻优,最终得到最佳提取工艺。结果皂苷和总黄酮分别在质量浓度0.008~0.056 g/L和0.024~0.080 g/L与吸光度具有良好的线性关系,方法学考察符合测定要求。选取隐层神经元个数为5的神经网络模型,优化遗传算法的各项参数后对甘草中皂苷和总黄酮进行提取工艺的目标优化,最终得到的最佳提取条件为:氨浓度0.62%,乙醇浓度64%,回流时间1.8 h,液固比12∶1。模型综合评价预测值为191.65,而按照上述最佳提取条件试验所得的平均综合评价值为188.90,两者相对误差为1.43%,证明神经网络和遗传算法具有较好的预测性。结论建立的数学模型寻求同时提取甘草皂苷和总黄酮最佳提取条件是科学可行的,为实现中药化学成分乃至药效物质基础多目标寻优提供了新的参考和思路。

Objective To optimize the simultaneous extracting technique of saponins and total flavonoids from licorice（ gancao）. Methods Ammonia concentration（ A）,ethanol concentration（ B）,reflux time（ C）,and liquid/solid ratio（ D） were set as the independent variables in this single factor experiment.Four factors and five levels of central composite design（ CCD） in response surface methods were used to determine the content of saponins and total flavonoids in licorice. This study used ultraviolet spectrophotometric method to measure saponins and total flavonoids in licorice at the wave length of 252 nm and 510 nm respectively. The entropy weight method in the R language application was used to assign weight to the above two parameters. The three-layered model of BP neural network was established to test the effect of the number of hidden neurons（ size）. Finally,genetic algorithm was established to optimize theextraction techniques with real-coded program of R language. Results This method achieved the objective of testing requirements. There was a good linear relationship between saponins at 0. 008 - 0. 056g/L,total flavonoids at 0. 024 - 0. 08 g/L,and light absorbance. This method set the neural network model with five hidden layer neurons. After optimizing the parameters of genetic algorithm,the extraction process of saponins and total flavonoids from licorice was optimized. The final optimal parameters were0. 62% ammonia,64% ethanol,1. 8 h reflux time,and 12 ∶ 1 of liquid-solid ratio. In this optimal extraction condition,predictive value of this model was 191. 65,and experimental average value was188. 90. The relative error was 1. 43%,which demonstrated a good predictability of the neural network model and genetic algorithm. Conclusion This mathematical model to optimize the extraction techniques of saponins and total flavonoids from licorice is scientific and feasible. It also provids an innovative reference and approach to the multi-objective extraction techniques for identifying chemical compound and active ingredients of traditional Chinese medicine.